Author: Uri Blass
Date: 02:18:22 02/01/04
Go up one level in this thread
On February 01, 2004 at 04:38:12, Tord Romstad wrote: >On January 31, 2004 at 13:43:39, Uri Blass wrote: > >>Note that evaluation that is good enough to win KRP vs KR does not have to be >>good enough to detect draws and if it only good enough to give wins better score >>than draws and simpler win better score than harder win then it is good enough >>when you already have the endgame but not always good enough before it. > >Yes. This is a really hard problem, and I haven't found a good way to deal with >it. For >just a single endgame like KRPKR it isn't very hard to tune the scores to be >reasonably >consistent with the evaluation of more complicated endgames, but when you add >code >for more and more classes of endgames it becomes very complicated to make the >program handle the transitions between different endgames correctly. > >If anyone has found a good way to handle this problem, I would be happy to learn >about it. > >Tord I think that first I need to add scores for positions when I can be sure in the final result based on evaluation. Movei today even does not know to evaluate KP vs K correctly and I probably should first teach it to evaluate some basic draws in KP vs K. examples: if the pawn is not in the 7th and the kin is one square behind the pawn it is a draw and the program should return 0.00 and stop to search in kp vs K if the pawn is unstoppable it is a win for white and the program can add mate bonus to the normal score and not search the position unless alpha is above mate bonus (I cannot see a case of stalemate because the stronger side always can promote the pawn to a rook). KP vs K unlike KRP vs KR are probably less important if the program support tablebases because when KP vs K is a leaf position already there is enough depth in tablebase position some plies before it but it may be a good exercise for more complicated problems. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.